A Robust Image Super-Resolution Scheme Based on Redescending M-Estimators and Information-Theoretic Divergence

This paper proposes a novel image super-resolution (SR) algorithm in a robust estimation framework. SR estimation is formulated as an optimization (minimization) problem whose objective function is based on robust M-estimators and its solution yields the SR output. The novelty of the proposed scheme lies in the selection of this class of estimators and the incorporation of information-theoretic similarity measures. Such a choice helps in dealing with violations (outliers) of the assumed mathematical model that generated the low-resolution images from the "unknown" high-resolution one. The proposed approach results in high-resolution images with no estimation artifacts. Experimental results demonstrate its superior performance in comparison to both L₁ and L₂ estimation in terms of robustness and speed of convergence.